diff options
Diffstat (limited to 'src/audio_dec/xine_dts_decoder.c')
-rw-r--r-- | src/audio_dec/xine_dts_decoder.c | 590 |
1 files changed, 590 insertions, 0 deletions
diff --git a/src/audio_dec/xine_dts_decoder.c b/src/audio_dec/xine_dts_decoder.c new file mode 100644 index 000000000..ffacf4a94 --- /dev/null +++ b/src/audio_dec/xine_dts_decoder.c @@ -0,0 +1,590 @@ +/* + * Copyright (C) 2000-2007 the xine project + * + * This file is part of xine, a unix video player. + * + * xine is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * xine is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110, USA + */ + +/** + * @file + * @brief DTS decoder for xine + * + * @author Joachim Koenig (2001-09-04) + * @author James Courtier-Dutton (2001-12-09) + */ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#ifndef __sun +/* required for swab() */ +#define _XOPEN_SOURCE 500 +#endif +/* avoid compiler warnings */ +#define _BSD_SOURCE 1 + +#include <stdlib.h> +#include <unistd.h> +#include <string.h> +#include <sys/types.h> +#include <sys/stat.h> +#include <fcntl.h> +#include <assert.h> + +#define LOG_MODULE "libdts" +#define LOG_VERBOSE +/* +#define LOG +*/ + +#include <xine/xine_internal.h> +#include <xine/xineutils.h> +#include <xine/audio_out.h> +#include <xine/buffer.h> + +#include "bswap.h" + +#include <dts.h> + +#define MAX_AC5_FRAME 4096 + +typedef struct { + audio_decoder_class_t decoder_class; +} dts_class_t; + +typedef struct { + audio_decoder_t audio_decoder; + + xine_stream_t *stream; + audio_decoder_class_t *class; + + dts_state_t *dts_state; + int64_t pts; + + int audio_caps; + int sync_state; + int ac5_length, ac5_pcm_length, frame_todo; + uint32_t syncdword; + uint8_t frame_buffer[MAX_AC5_FRAME]; + uint8_t *frame_ptr; + + int output_open; + + int bypass_mode; + int dts_flags; + int dts_sample_rate; + int dts_bit_rate; + int dts_flags_map[11]; /* Convert from stream dts_flags to the dts_flags we want from the dts downmixer */ + int ao_flags_map[11]; /* Convert from the xine AO_CAP's to dts_flags. */ + int have_lfe; + + +} dts_decoder_t; + +static void dts_reset (audio_decoder_t *const this_gen) { +} + +static void dts_discontinuity (audio_decoder_t *const this_gen) { +} + +/** + * @brief Convert a array of floating point samples into 16-bit signed integer samples + * @param f Floating point samples array (origin) + * @param s16 16-bit signed integer samples array (destination) + * @param num_channels Number of channels present in the stream + * + * @todo This same work is being done in many decoders to adapt the output of + * the decoder to what the audio output can actually use, this should be + * done by the audio_output loop, not by the decoders. + * @note This is subtly different from the function with the same name in xine_musepack_decoder.c + */ +static inline void float_to_int (const float *const _f, int16_t *const s16, const int num_channels) { + const int endidx = 256 * num_channels; + int i, j; + + for (i = 0, j = 0; j < endidx; i++, j += num_channels) { + const float f = _f[i] * 32767; + if (f > INT16_MAX) + s16[j] = INT16_MAX; + else if (f < INT16_MIN) + s16[j] = INT16_MIN; + else + s16[j] = f; + /* printf("samples[%d] = %f, %d\n", i, _f[i], s16[num_channels*i]); */ + } +} + +static inline void mute_channel (int16_t *const s16, const int num_channels) { + const int endidx = 256 * num_channels; + int i; + + for (i = 0; i < endidx; i += num_channels) + s16[i] = 0; +} + +static void dts_decode_frame (dts_decoder_t *this, const int64_t pts) { + + audio_buffer_t *audio_buffer; + uint32_t ac5_spdif_type=0; + int output_mode = AO_CAP_MODE_STEREO; + uint8_t *data_out; + uint8_t *const data_in = this->frame_buffer; + + lprintf("decode_frame\n"); + audio_buffer = this->stream->audio_out->get_buffer (this->stream->audio_out); + audio_buffer->vpts = pts; + + if(this->bypass_mode) { + /* SPDIF digital output */ + if (!this->output_open) { + this->output_open = ((this->stream->audio_out->open) (this->stream->audio_out, this->stream, + 16, this->dts_sample_rate, + AO_CAP_MODE_AC5)); + } + + if (!this->output_open) + return; + + data_out=(uint8_t *) audio_buffer->mem; + if (this->ac5_length > 8191) { + xprintf(this->stream->xine, XINE_VERBOSITY_DEBUG, "libdts: ac5_length too long\n"); + this->ac5_pcm_length = 0; + } + + switch (this->ac5_pcm_length) { + case 512: + ac5_spdif_type = 0x0b; /* DTS-1 (512-sample bursts) */ + break; + case 1024: + ac5_spdif_type = 0x0c; /* DTS-1 (1024-sample bursts) */ + break; + case 2048: + ac5_spdif_type = 0x0d; /* DTS-1 (2048-sample bursts) */ + break; + default: + xprintf(this->stream->xine, XINE_VERBOSITY_DEBUG, + "libdts: DTS %i-sample bursts not supported\n", this->ac5_pcm_length); + return; + } + +#ifdef LOG_DEBUG + { + int i; + printf("libdts: DTS frame type=%d\n",data_in[4] >> 7); + printf("libdts: DTS deficit frame count=%d\n",(data_in[4] & 0x7f) >> 2); + printf("libdts: DTS AC5 PCM samples=%d\n",ac5_pcm_samples); + printf("libdts: DTS AC5 length=%d\n",this->ac5_length); + printf("libdts: DTS AC5 bitrate=%d\n",((data_in[8] & 0x03) << 4) | (data_in[8] >> 4)); + printf("libdts: DTS AC5 spdif type=%d\n", ac5_spdif_type); + + printf("libdts: "); + for(i=2000;i<2048;i++) { + printf("%02x ",data_in[i]); + } + printf("\n"); + } +#endif + + lprintf("length=%d pts=%"PRId64"\n",this->ac5_pcm_length,audio_buffer->vpts); + + audio_buffer->num_frames = this->ac5_pcm_length; + + // Checking if AC5 data plus IEC958 header will fit into frames samples data + if ( this->ac5_length + 8 <= this->ac5_pcm_length * 2 * 2 ) { + data_out[0] = 0x72; data_out[1] = 0xf8; /* spdif syncword */ + data_out[2] = 0x1f; data_out[3] = 0x4e; /* .............. */ + data_out[4] = ac5_spdif_type; /* DTS data */ + data_out[5] = 0; /* Unknown */ + data_out[6] = (this->ac5_length << 3) & 0xff; /* ac5_length * 8 */ + data_out[7] = ((this->ac5_length ) >> 5) & 0xff; + + if( this->ac5_pcm_length ) { + if( this->ac5_pcm_length % 2) { + swab(data_in, &data_out[8], this->ac5_length ); + } else { + swab(data_in, &data_out[8], this->ac5_length + 1); + } + } + // Transmit it without header otherwise, receivers will autodetect DTS + } else { + lprintf("AC5 data is too large (%i > %i), sending without IEC958 header\n", + this->ac5_length + 8, this->ac5_pcm_length * 2 * 2); + memcpy(data_out, data_in, this->ac5_length); + } + } else { + /* Software decode */ + int i, dts_output_flags; + int16_t *const int_samples = audio_buffer->mem; + int number_of_dts_blocks; + + level_t level = 1.0; + sample_t *samples; + + dts_output_flags = this->dts_flags_map[this->dts_flags & DTS_CHANNEL_MASK]; + + if(dts_frame(this->dts_state, data_in, &dts_output_flags, &level, 0)) { + xprintf (this->stream->xine, XINE_VERBOSITY_DEBUG, "libdts: dts_frame error\n"); + return; + } + + this->have_lfe = dts_output_flags & DTS_LFE; + if (this->have_lfe) + if (this->audio_caps & AO_CAP_MODE_5_1CHANNEL) { + output_mode = AO_CAP_MODE_5_1CHANNEL; + } else if (this->audio_caps & AO_CAP_MODE_4_1CHANNEL) { + output_mode = AO_CAP_MODE_4_1CHANNEL; + } else { + xprintf(this->stream->xine, XINE_VERBOSITY_DEBUG, "libdts: WHAT DO I DO!!!\n"); + output_mode = this->ao_flags_map[dts_output_flags & DTS_CHANNEL_MASK]; + } + else + output_mode = this->ao_flags_map[dts_output_flags & DTS_CHANNEL_MASK]; + + if (!this->output_open) { + this->output_open = (this->stream->audio_out->open) (this->stream->audio_out, this->stream, + 16, this->dts_sample_rate, + output_mode); + } + + if (!this->output_open) + return; + number_of_dts_blocks = dts_blocks_num (this->dts_state); + audio_buffer->num_frames = 256*number_of_dts_blocks; + for(i = 0; i < number_of_dts_blocks; i++) { + if(dts_block(this->dts_state)) { + xprintf(this->stream->xine, XINE_VERBOSITY_DEBUG, + "libdts: dts_block error on audio channel %d\n", i); + audio_buffer->num_frames = 0; + break; + } + + samples = dts_samples(this->dts_state); + switch (output_mode) { + case AO_CAP_MODE_MONO: + float_to_int (&samples[0], int_samples+(i*256), 1); + break; + case AO_CAP_MODE_STEREO: + /* Tested, working. */ + float_to_int (&samples[0*256], int_samples+(i*256*2), 2); /* L */ + float_to_int (&samples[1*256], int_samples+(i*256*2)+1, 2); /* R */ + break; + case AO_CAP_MODE_4CHANNEL: + /* Tested, working */ + float_to_int (&samples[0*256], int_samples+(i*256*4), 4); /* L */ + float_to_int (&samples[1*256], int_samples+(i*256*4)+1, 4); /* R */ + float_to_int (&samples[2*256], int_samples+(i*256*4)+2, 4); /* RL */ + float_to_int (&samples[3*256], int_samples+(i*256*4)+3, 4); /* RR */ + break; + case AO_CAP_MODE_4_1CHANNEL: + /* Tested, working */ + float_to_int (&samples[0*256], int_samples+(i*256*6)+0, 6); /* L */ + float_to_int (&samples[1*256], int_samples+(i*256*6)+1, 6); /* R */ + float_to_int (&samples[2*256], int_samples+(i*256*6)+2, 6); /* RL */ + float_to_int (&samples[3*256], int_samples+(i*256*6)+3, 6); /* RR */ + float_to_int (&samples[4*256], int_samples+(i*256*6)+5, 6); /* LFE */ + mute_channel ( int_samples+(i*256*6)+4, 6); /* C */ + break; + case AO_CAP_MODE_5CHANNEL: + /* Tested, working */ + float_to_int (&samples[0*256], int_samples+(i*256*6)+4, 6); /* C */ + float_to_int (&samples[1*256], int_samples+(i*256*6)+0, 6); /* L */ + float_to_int (&samples[2*256], int_samples+(i*256*6)+1, 6); /* R */ + float_to_int (&samples[3*256], int_samples+(i*256*6)+2, 6); /* RL */ + float_to_int (&samples[4*256], int_samples+(i*256*6)+3, 6); /* RR */ + mute_channel ( int_samples+(i*256*6)+5, 6); /* LFE */ + break; + case AO_CAP_MODE_5_1CHANNEL: + float_to_int (&samples[0*256], int_samples+(i*256*6)+4, 6); /* C */ + float_to_int (&samples[1*256], int_samples+(i*256*6)+0, 6); /* L */ + float_to_int (&samples[2*256], int_samples+(i*256*6)+1, 6); /* R */ + float_to_int (&samples[3*256], int_samples+(i*256*6)+2, 6); /* RL */ + float_to_int (&samples[4*256], int_samples+(i*256*6)+3, 6); /* RR */ + float_to_int (&samples[5*256], int_samples+(i*256*6)+5, 6); /* LFE */ /* Not working yet */ + break; + default: + xprintf (this->stream->xine, XINE_VERBOSITY_DEBUG, "libdts: help - unsupported mode %08x\n", output_mode); + } + } + } + + this->stream->audio_out->put_buffer (this->stream->audio_out, audio_buffer, this->stream); + + +} + +static void dts_decode_data (audio_decoder_t *this_gen, buf_element_t *buf) { + + dts_decoder_t *const this = (dts_decoder_t *) this_gen; + uint8_t *current = (uint8_t *)buf->content; + uint8_t *sync_start=current + 1; + uint8_t *const end = buf->content + buf->size; + + lprintf("decode_data\n"); + + if (buf->decoder_flags & BUF_FLAG_PREVIEW) + return; + if (buf->decoder_flags & BUF_FLAG_STDHEADER) + return; + + lprintf ("processing...state %d\n", this->sync_state); + + while (current < end) { + switch (this->sync_state) { + case 0: /* Looking for sync header */ + this->syncdword = (this->syncdword << 8) | *current++; +/* + if ((this->syncdword == 0xff1f00e8) || + (this->syncdword == 0x1fffe800) || + (this->syncdword == 0xfe7f0180) || + (this->syncdword == 0x7ffe8001) ) { +*/ + + if ((this->syncdword == 0x7ffe8001) || (this->syncdword == 0xff1f00e8)) { + const uint32_t be_syncdword = be2me_32(this->syncdword); + + lprintf ("sync found: syncdword=0x%x\n", this->syncdword); + + memcpy(this->frame_buffer, &be_syncdword, sizeof(be_syncdword)); + + this->sync_state = 1; + this->frame_ptr = this->frame_buffer+4; + this->pts = buf->pts; + } + break; + + case 1: /* Looking for enough bytes for sync_info. */ + sync_start = current - 1; + *this->frame_ptr++ = *current++; + if ((this->frame_ptr - this->frame_buffer) > 19) { + const int old_dts_flags = this->dts_flags; + const int old_dts_sample_rate = this->dts_sample_rate; + const int old_dts_bit_rate = this->dts_bit_rate; + + this->ac5_length = dts_syncinfo (this->dts_state, this->frame_buffer, + &this->dts_flags, + &this->dts_sample_rate, + &this->dts_bit_rate, &(this->ac5_pcm_length)); + lprintf("ac5_length=%d\n",this->ac5_length); + lprintf("dts_sample_rate=%d\n",this->dts_sample_rate); + + if ( (this->ac5_length < 80) || (this->ac5_length > MAX_AC5_FRAME) ) { /* Invalid dts ac5_pcm_length */ + this->syncdword = 0; + current = sync_start; + this->sync_state = 0; + break; + } + + lprintf("Frame length = %d\n",this->ac5_pcm_length); + + this->frame_todo = this->ac5_length - 20; + this->sync_state = 2; + if (!_x_meta_info_get(this->stream, XINE_META_INFO_AUDIOCODEC) || + old_dts_flags != this->dts_flags || + old_dts_sample_rate != this->dts_sample_rate || + old_dts_bit_rate != this->dts_bit_rate) { + + switch (this->dts_flags & DTS_CHANNEL_MASK) { + case DTS_3F2R: + if (this->dts_flags & DTS_LFE) + _x_meta_info_set_utf8(this->stream, XINE_META_INFO_AUDIOCODEC, "DTS 5.1"); + else + _x_meta_info_set_utf8(this->stream, XINE_META_INFO_AUDIOCODEC, "DTS 5.0"); + break; + case DTS_3F1R: + case DTS_2F2R: + if (this->dts_flags & DTS_LFE) + _x_meta_info_set_utf8(this->stream, XINE_META_INFO_AUDIOCODEC, "DTS 4.1"); + else + _x_meta_info_set_utf8(this->stream, XINE_META_INFO_AUDIOCODEC, "DTS 4.0"); + break; + case DTS_2F1R: + case DTS_3F: + _x_meta_info_set_utf8(this->stream, XINE_META_INFO_AUDIOCODEC, "DTS 3.0"); + break; + case DTS_STEREO: + _x_meta_info_set_utf8(this->stream, XINE_META_INFO_AUDIOCODEC, "DTS 2.0 (stereo)"); + break; + case DTS_MONO: + _x_meta_info_set_utf8(this->stream, XINE_META_INFO_AUDIOCODEC, "DTS 1.0"); + break; + default: + _x_meta_info_set_utf8(this->stream, XINE_META_INFO_AUDIOCODEC, "DTS"); + break; + } + + _x_stream_info_set(this->stream, XINE_STREAM_INFO_AUDIO_BITRATE, this->dts_bit_rate); + _x_stream_info_set(this->stream, XINE_STREAM_INFO_AUDIO_SAMPLERATE, this->dts_sample_rate); + } + } + break; + + case 2: /* Filling frame_buffer with sync_info bytes */ + *this->frame_ptr++ = *current++; + this->frame_todo--; + if (this->frame_todo < 1) { + this->sync_state = 3; + } else break; + + case 3: /* Ready for decode */ +#if 0 + dtsdec_decode_frame (this, this->pts_list[0]); +#else + dts_decode_frame (this, this->pts); +#endif + case 4: /* Clear up ready for next frame */ + this->pts = 0; + this->syncdword = 0; + this->sync_state = 0; + break; + default: /* No come here */ + break; + } + } +} + +static void dts_dispose (audio_decoder_t *this_gen) { + dts_decoder_t *const this = (dts_decoder_t *) this_gen; + + if (this->output_open) + this->stream->audio_out->close (this->stream->audio_out, this->stream); + + free (this); +} + +static audio_decoder_t *open_plugin (audio_decoder_class_t *class_gen, xine_stream_t *stream) { + dts_decoder_t *this ; + + lprintf("open_plugin\n"); + + this = calloc(1, sizeof (dts_decoder_t)); + + this->audio_decoder.decode_data = dts_decode_data; + this->audio_decoder.reset = dts_reset; + this->audio_decoder.discontinuity = dts_discontinuity; + this->audio_decoder.dispose = dts_dispose; + + this->dts_state = dts_init(0); + this->audio_caps = stream->audio_out->get_capabilities(stream->audio_out); + if(this->audio_caps & AO_CAP_MODE_AC5) + this->bypass_mode = 1; + else { + this->bypass_mode = 0; + /* FIXME: Leave "DOLBY pro logic" downmix out for now. */ + this->dts_flags_map[DTS_MONO] = DTS_MONO; + this->dts_flags_map[DTS_STEREO] = DTS_STEREO; + this->dts_flags_map[DTS_3F] = DTS_STEREO; + this->dts_flags_map[DTS_2F1R] = DTS_STEREO; + this->dts_flags_map[DTS_3F1R] = DTS_STEREO; + this->dts_flags_map[DTS_2F2R] = DTS_STEREO; + this->dts_flags_map[DTS_3F2R] = DTS_STEREO; + + this->ao_flags_map[DTS_MONO] = AO_CAP_MODE_MONO; + this->ao_flags_map[DTS_STEREO] = AO_CAP_MODE_STEREO; + this->ao_flags_map[DTS_3F] = AO_CAP_MODE_STEREO; + this->ao_flags_map[DTS_2F1R] = AO_CAP_MODE_STEREO; + this->ao_flags_map[DTS_3F1R] = AO_CAP_MODE_STEREO; + this->ao_flags_map[DTS_2F2R] = AO_CAP_MODE_STEREO; + this->ao_flags_map[DTS_3F2R] = AO_CAP_MODE_STEREO; + + /* find best mode */ + if (this->audio_caps & AO_CAP_MODE_5_1CHANNEL) { + + this->dts_flags_map[DTS_2F2R] = DTS_2F2R; + this->dts_flags_map[DTS_3F2R] = DTS_3F2R | DTS_LFE; + this->ao_flags_map[DTS_2F2R] = AO_CAP_MODE_4CHANNEL; + this->ao_flags_map[DTS_3F2R] = AO_CAP_MODE_5CHANNEL; + + } else if (this->audio_caps & AO_CAP_MODE_5CHANNEL) { + + this->dts_flags_map[DTS_2F2R] = DTS_2F2R; + this->dts_flags_map[DTS_3F2R] = DTS_3F2R; + this->ao_flags_map[DTS_2F2R] = AO_CAP_MODE_4CHANNEL; + this->ao_flags_map[DTS_3F2R] = AO_CAP_MODE_5CHANNEL; + + } else if (this->audio_caps & AO_CAP_MODE_4_1CHANNEL) { + + this->dts_flags_map[DTS_2F2R] = DTS_2F2R; + this->dts_flags_map[DTS_3F2R] = DTS_2F2R | DTS_LFE; + this->ao_flags_map[DTS_2F2R] = AO_CAP_MODE_4CHANNEL; + this->ao_flags_map[DTS_3F2R] = AO_CAP_MODE_4CHANNEL; + + } else if (this->audio_caps & AO_CAP_MODE_4CHANNEL) { + + this->dts_flags_map[DTS_2F2R] = DTS_2F2R; + this->dts_flags_map[DTS_3F2R] = DTS_2F2R; + + this->ao_flags_map[DTS_2F2R] = AO_CAP_MODE_4CHANNEL; + this->ao_flags_map[DTS_3F2R] = AO_CAP_MODE_4CHANNEL; + + /* else if (this->audio_caps & AO_CAP_MODE_STEREO) + defaults are ok */ + } else if (!(this->audio_caps & AO_CAP_MODE_STEREO)) { + xprintf (this->stream->xine, XINE_VERBOSITY_LOG, _("HELP! a mono-only audio driver?!\n")); + + this->dts_flags_map[DTS_MONO] = DTS_MONO; + this->dts_flags_map[DTS_STEREO] = DTS_MONO; + this->dts_flags_map[DTS_3F] = DTS_MONO; + this->dts_flags_map[DTS_2F1R] = DTS_MONO; + this->dts_flags_map[DTS_3F1R] = DTS_MONO; + this->dts_flags_map[DTS_2F2R] = DTS_MONO; + this->dts_flags_map[DTS_3F2R] = DTS_MONO; + + this->ao_flags_map[DTS_MONO] = AO_CAP_MODE_MONO; + this->ao_flags_map[DTS_STEREO] = AO_CAP_MODE_MONO; + this->ao_flags_map[DTS_3F] = AO_CAP_MODE_MONO; + this->ao_flags_map[DTS_2F1R] = AO_CAP_MODE_MONO; + this->ao_flags_map[DTS_3F1R] = AO_CAP_MODE_MONO; + this->ao_flags_map[DTS_2F2R] = AO_CAP_MODE_MONO; + this->ao_flags_map[DTS_3F2R] = AO_CAP_MODE_MONO; + } + } + this->stream = stream; + this->class = class_gen; + this->output_open = 0; + + return &this->audio_decoder; +} + +static void *init_plugin (xine_t *xine, void *data) { + dts_class_t *this ; + + lprintf("init_plugin\n"); + + this = calloc(1, sizeof (dts_class_t)); + + this->decoder_class.open_plugin = open_plugin; + this->decoder_class.identifier = "DTS"; + this->decoder_class.description = N_("DTS passthru audio format decoder plugin"); + this->decoder_class.dispose = default_audio_decoder_class_dispose; + + return this; +} + +static const uint32_t audio_types[] = { + BUF_AUDIO_DTS, 0 + }; + +static const decoder_info_t dec_info_audio = { + audio_types, /* supported types */ + 1 /* priority */ +}; + +const plugin_info_t xine_plugin_info[] EXPORTED = { + /* type, API, "name", version, special_info, init_function */ + { PLUGIN_AUDIO_DECODER, 16, "dts", XINE_VERSION_CODE, &dec_info_audio, init_plugin }, + { PLUGIN_NONE, 0, "", 0, NULL, NULL } +}; |